circle_bundles.FrameReducerConfig

class circle_bundles.FrameReducerConfig(method='none', stage='pre_classifying', d=0, max_frames=None, rng_seed=0, psc_verbosity=0)[source]

Bases: object

Configuration for O(2)-equivariant dimensionality reduction of Stiefel frames.

Parameters:

method (Literal['none', 'subspace_pca', 'psc'])
stage (Literal['pre_classifying', 'post_stiefel'])
d (int)
max_frames (int | None)
rng_seed (int)
psc_verbosity (int)

method

“none” : no reduction
“subspace_pca”: always-available, O(2)-equivariant baseline
“psc” : PSC-based reducer (optional dependency)

Type:: ReduceMethod

stage

“pre_classifying”: reduce the raw Stiefel point cloud before building the classifying map
“post_stiefel” : reduce after a Stiefel projection step (legacy / experimental)

Type:: ReduceStage

d

Target ambient dimension after reduction (must satisfy 2 <= d <= D at runtime). A value of 0 is treated as “unset” by callers (i.e., no reduction unless set).

Type:: int

max_frames

Optional subsampling cap for fitting the reducer (speed control).

Type:: Optional[int]

rng_seed

RNG seed used for subsampling.

Type:: int

psc_verbosity

Verbosity forwarded to PSC’s optimization routine (if method=”psc”).

Type:: int

__init__(method='none', stage='pre_classifying', d=0, max_frames=None, rng_seed=0, psc_verbosity=0)

Parameters:

method (Literal['none', 'subspace_pca', 'psc'])
stage (Literal['pre_classifying', 'post_stiefel'])
d (int)
max_frames (int | None)
rng_seed (int)
psc_verbosity (int)

Return type:

None

Methods

__init__([method, stage, d, max_frames, ...])

Attributes

`d`
`max_frames`
`method`
`psc_verbosity`
`rng_seed`
`stage`